Original · Unique Research · 2026-03-18
Editor's note: This is a complete translation of the published March 18 article, whose source says it edited and shortened the March 17 livestream; it is not an unabridged livestream transcript. Chen Peilin and Zhang Xiaolong are romanized names; Kelly is the source's supplied name. Employment history, layoffs, work-replacement percentages, messages sent, token consumption, follower growth and GitHub statistics are source or speaker claims, not independently audited outcomes. The source uses different time windows and measures for code commits and GitHub contributions; those figures are not treated as interchangeable. A company total of 50 Agents and Kelly's statement about personally reviewing 5 are retained at their stated scopes. The 700-person example and predictions that departments can be emptied out are not evidence of verified layoffs at that other company. Terms such as “species,” “eating people” and “motherly” are the source's rhetorical metaphors, not biological classifications or hiring requirements. The mass-messaging and account-access anecdotes do not establish recipient consent, security compliance or permission to reproduce those actions. Human confirmation, review exhaustion and dependency on offline Agents are retained. Claims about replacing software, the “remaining 98%” and Buda.im's future are historical opinions or ambitions, not established market facts or endorsements.
Unique Research · Late-Night Lobster Conversations
He Used 50 AIs and Laid Off 40% of His Company
A Former HEYTEA CTO's Real-World Experiment: Not Layoffs, but Work Being Consumed by AI
"
Last week, I even held a layoff meeting. This time, the layoffs were not people. They were lobsters.
Chen Peilin (Kelly)—founder of vika, HEYTEA's first CTO, and creator of the open-source APITable project (15,000 Stars)
His company manages 50 “lobsters”—OpenClaw AI Agents—that took over 90% of its daily work. In the second week after deploying them, the company laid off 40% of its people.
Kelly has not written a line of code since July 2025, yet has made more than 5000 code commits over the past 12 months.
His exact words were: “I am genuinely very worried. If business owners all realize the power of using lobsters in a company, society may face a wave of mass layoffs.”
From a Side Activity to the Main Force
Kelly's lobster-raising journey began with a “toy” in January 2026.
OpenClaw had just appeared then. He kept a single lobster to browse Twitter. Lying on a bench at Vancouver airport, he watched code run automatically, collect TikTok data and gain a few hundred followers.
"
At the time, it felt like a minor side activity.
The turning point came during the 2026 Spring Festival.
During the Lunar New Year's Eve meal, Kelly held chopsticks in his left hand and a foldable phone in his right, using AI to send 9000 WeChat messages.
"
I did not spend a cent on tokens. I even sent one to Zhang Xiaolong, who replied: “Was this sent by AI?”
Even so, the lobster was still a toy in his eyes. Its real incursion came on the first day back at work after the Spring Festival.
30 Groups in Two Days: AI Begins to Enter the Company
“Back from the Spring Festival, I needed to assign work. I suddenly thought: why not create a series of groups and bring the lobsters in?”
Kelly did not expect that idea to turn the company into a lobster empire within two days: 30 Discord groups, each with a lobster employee.
More unexpectedly, the first people to embrace AI were not programmers but administration, finance and HR.
"
They used to be reluctant to invest anything in AI, always saying it was useless or too stupid. This time, they were the pioneers.
Kelly analyzed the reasons:
• Administration handles invoices and documents: deterministic tasks.
• Finance handles bookkeeping and reconciliation: clear rules.
• Sales replies to emails and updates the CRM: highly repetitive work.
These are precisely the things AI does best.
"
I configured an administrative-document lobster and gave it our historical company files to manage. It was incredibly satisfying.
There is also a bookkeeping lobster: it automatically retrieves emailed invoices, recognizes and classifies them, and records the accounts. AI handles it all.
The cloud-cost lobster goes further: it scans cloud servers, identifies idle resources and directly saves the company money.
But what about the R&D department?
"
Our R&D colleagues use it the worst. They say, “Come on, isn't this just Cursor?”
Kelly smiled wryly: “Typical programmer thinking.”
40% Layoffs: When AI Starts Consuming Jobs
After raising 50 lobsters, Kelly discovered a frightening fact:
Once requirements can be described precisely, half of the human work is already done.
“I used to assign everyone tasks in weekly meetings. Now I find that if a colleague only accepts tasks each week and cannot identify requirements, AI will consume their entire daily workload.”
Ultimately, with AI having taken over so much work, the team temporarily had no work left to distribute, and the company was forced to lay off 40% of its staff.
"
Mostly R&D. When you can clearly describe the requirements, AI can finish in a day. Do you still need so many people?
He even changed the weekly meeting process. It is now called the AI managers' meeting.
"
Nobody needs to speak. Just put the prompts here. Once a prompt appears, the work magically gets done within a day. What is difficult? Finding the people who identify requirements and problems.
Which Category Are You In?
At this point, you may get the impression that the people laid off lacked ability.
But Kelly's conclusion is the opposite.
The people being replaced are not those with weak skills, but those who only know how to accept assigned tasks.
In his company, people are starting to fall into two categories:
• Those who wait each week to be assigned work;
• Those who ask themselves each week, “What should I have AI do?”
The first group is being replaced by AI.
The second is managing AI.
This is not a question of efficiency. It is a question of species.
Burning 700 Million Tokens a Day
Raising lobsters is not free.
Kelly disclosed a startling number: at the peak, usage exceeded 700 million tokens a day, later optimized to more than 200 million.
"
Quite expensive.
But Kelly thinks it is worth it.
“My GitHub now has 16000 contributions, about 10000 of which were submitted by lobsters. I recently stopped it because there were too many.”
"
If I have an idea and give an instruction before bed, it is there when I wake up the next morning.
How the 50 Lobsters Divide the Work
Kelly showed his company's organizational structure in Discord:
• General management: administration, finance and legal affairs.
• Sales: customer-email analysis and CRM updates.
• Marketing operations: social-media operations and content publishing.
• Content operations: topic selection, writing and formatting.
• R&D: creative development and operations development.
• Open-source-project operations lobster: automatically review GitHub Issues, suggest replies and even close them directly.
“The most satisfying one is the open-source-project operations lobster. Replying to Issues used to be painful. Nobody can know every product feature and every code detail. But AI can: it knows the codebase, the product logic and even the prices.”
He demonstrated one case: AI suggested a reply to an Issue; he answered “yes,” and it sent the reply automatically.
"
It is like reviewing memorials to the throne.
Kelly described his current situation: “Handling countless affairs every day, I finally understand what it feels like to be an ineffectual emperor. Discord keeps pinging, and I cannot read all these memorials. Help!”
"
I am exhausted on my own. I cannot review it all.
More alarmingly, when a lobster goes offline, employees react like this:
"
My lobster is going offline—come and help me right away! I need the lobster; I cannot find the files!
Kelly observed: “Apparently, a lobster taking leave is no less serious than the company losing power.”
A Frightening Story: In a Company of 700, 680 Roles Might Be Emptied Out
Kelly shared a case that left him thinking for a long time.
He told a CTO about three remedies:
1. An operations-log lobster: inspect logs daily, proactively fix an Error and create a PR.
2. A requirements-handling lobster: scan requirements and Bug tables daily and process them automatically.
3. A testing-coordination lobster: tag the testers directly in the PR.
“Your company has 700 people in R&D, but only 20 in testing.”
Kelly asked him: “What do the remaining 680 do?”
The answer left the CTO silent: “All their work is emptied out.”
"
AI consumes everything and moves the work to testing. Testing only needs to verify.
Kelly's conclusion is that the R&D programming department of a mature software company can have its work emptied out.
"
Including our own company.
This is not a hypothetical.
It is a future that can already be tested.
AI Is Eating Software
Kelly offered a more radical view:
AI is eating software.
“Previously, we instinctively thought: Build—write code, create UI and make tools. But people without that baggage do not want a UI, a system or a tool. They want their problem solved.”
He gave an example:
A customer wanted an e-commerce operations system to monitor advertising traffic. Kelly asked: “Why not just ask the lobster about yesterday's advertising traffic?”
The customer paused: “Oh, right.”
"
So I even think Vibe Coding is becoming outdated. When a problem can be solved with a single Skill, traditional UI software may lose its purpose, and traditional UI software is exactly what Vibe Coding produces.
Administration does not need an interface to complete tasks. Sales does not need one to reply to emails. Finance does not need one for bookkeeping.
"
The era has arrived when business staff can free themselves completely from programmers and product managers and solve problems directly without a UI.
Where Does This Leave People?
Kelly did not let fear of AI replacing humanity carry him away. He offered his own answer:
People are still needed, but the requirements for talent have changed.
He summed up his new talent ideal as motherly, benevolent people:
• A mother's heart: genuinely care about a lobster's development and growth rather than neglecting it.
• A heart inclined toward good: know what is good and use technology for good.
• Responsibility and Ownership: humans determine a lobster's ceiling.
"
I am already exhausted managing 5 lobsters and cannot review everything. The company still needs people, but their role has changed from executors to managers, questioners and decision-makers.
Kelly now sits at his computer for more than 12 hours a day, mostly interacting with AI.
"
Unless I am going out to meet people or using WeChat, I am basically talking to AI.
I talk to AI first, then tag a person to confirm.
The Company of the Future
Kelly is working on a new product, Buda.im.
"
I cannot deliver a Mac Mini to everyone, deploy OpenClaw and install Skills. I want them to open a website and hire an entire team within three minutes.
His vision is to turn the lobsters into a company.
"
What we should ultimately sell is an entire company, not a single lobster.
Kelly believes this is the right approach for the remaining 98%: people who cannot use Cursor or open a terminal, or even use a mouse.
"
The mass market does not care whether you are OpenClaw or something else. It cares whether you can hire 10 employees for it within a day and get them working.
Closing Thoughts
At the end of his talk, Kelly said:
"
I think the basic assumptions of economics and management have changed. For more than 100 years of management history, the underlying assumption has been how an organization can organize people effectively. That assumption is gradually collapsing.
Everything is renewing, and industries are reviving. This is only the beginning. The form of the company, management and the economy have changed. Everything needs to be rethought.
He set himself a Mission: protect humanity and advance human progress.
“Once most founders learn about this,”
“the entire historical moment will change.”
This article was compiled from the March 17, 2026 Late-Night Lobster Conversations livestream, with cuts and editing in the original source.